It’s evident now that tomorrow’s data centers are going to need a lot of power to operate.
I’ve been seeing these very sudden initiatives toward safe and common U.S. nuclear power, with Microsoft and Constellation reopening Three Mile Island, or groundbreaking work on small local nuclear plants.
There’s also quite a bit of research on renewables and how that would work. I even wrote about these new plans to generate power and perform AI operations on satellites orbiting the Earth,in order to harvest power in a different way.
But let’s look at some of the underlying challenges and our history of solving similar problems.
Electrical Engineering as a Guide
In a recent presentation, Deborah Douglas points out that as we go about solving the AI power problem, we already have some historical examples that can inform our search. Douglas is Senior Director of Collections and Curator of Science and Technology at the MIT Museum.
She talks about early electrical systems, and how the grid evolved from disparate networks.
I would also add that you can make the analogy to the Internet as well. You can talk about the decentralized Internet, too. But another thing that Douglas mentions is the actual tools that professionals used for engineering well before our modern electrical systems were the norm.
(To clarify: of course there was electricity, it just wasn’t in the modern systems and formats that we see now.)
Douglas takes us back to the 1940s, where there were lots of changes going on – not just changes to power grids, but in society, too.
One Woman’s Tale
She mentions the example of Phyllis Fox, who was working on a master’s thesis right here at MIT and submitted it in 1949. The title? “The solution of power network problems on large scale digital computers.”
It’s a story of the glass ceiling, and tenacity, and the power to change the world.
Douglas also mentions Vannevar Bush, who’s popular at MIT, too, and taught electrical engineering in 1919. He had a machine called the differential analyzer, and Fox was interested in it.
A Hard Day’s Work
Douglas takes us through a sort of visual picture of how Foxworked at the GE company as an engineering assistant.
“She was put in an office with a Marchant calculator and given sets of problems to calculate, on average, she did one equation per minute. So 60 per hour, you know, 300, 400 per day, if you will. And they were expected to stay at it. And there was a minder, a man who walked up and down to make sure that they were doing their work at the time, and no talking was allowed. Well, Phyllis was one of these adventurous souls, and she wandered around the campus and she found during her lunch hour there was an office on the lower level that had, a calculator.And, they were doing the solutions by hand upstairs. So she went down in her lunch hour and did a full day’s work on the calculator. and then she would go back upstairs and pretend to be working, very clever in my estimation. But she also, in those meanderings, discovered that GE had purchased a differential analyzer, just like the one that I just showed you in the picture from MIT.”
Eventually, she notes, Fox got to work on the differential analyzer.
That calculator thing reminds me of what I was writing about the other day, with MIT professor Ethan Mollick‘s idea of the ‘wait calculation’ – that you don’t need to do all kinds of hard manual work, if you’re able to wait for modern tools to become available to automate that same work.
Anyway, in this case, her adventurous nature was rewarded, and she got to participate in the evolution of a pretty important field.
However, Douglas notes that Fox then got fired when the men came back from World War II – and then pursued that master’s degree.
Then she found another employer.
His name, Douglas adds, was Jay Forrester.
“He was ecumenical in his hiring,” she says. “So he hired women, he hired men. He hired the first black computer operator in the United States. He hired a blind computer operator. He hired people of Japanese descent, which was very controversial in the middle of World War II. He embraced (hiring) Phyllis Fox, who he thought was very bright.”
So what did these pioneers do with the differential analyzers?
Some Big Machines
I went to the MIT library site and took a look at some of those early machines, including something called the Rockefeller differential analyzer.
These are behemoths the size of washer-dryer systems, that have big gears and wheels that will do the calculations on a differential equation.
They’re really inspiring examples of analog computing.
Anyway, eventually Fox ended up teaching over at MIT, and showing more about how process calculations work, using flow charts, block diagrams, and that kind of thing.
She got a job with the Atomic Energy Commission’s Computer Center, and eventually developed something called Dynamo, as early computer simulation language. Fox also reportedly wrote the first LISP manual, worked at the Newark College of Engineering (now the Jersey Institute of Technology) and gained tenure in 1972.
It’s all incredibly impressive, and Douglas ended her presentation by theorizing that maybe there’s someone around today who will play this kind of role in solving our current bottlenecks and challenges.
I, for one, certainly hope so. But just thinking about the history of differential analyzer and analog equipment shows me that we have the ability to create new tools to do what we need them to do in pioneering AI power solutions.
In other words, it might sound intimidating to think that data centers will need X gigawatts and terawatts of power. But who knows how we’re going to view this in 10 or 20 years? Or even sooner?
Think about these inspirational examples, and do yourself a favor – go back and look at some of these now-antiquated machines. Consider that at their time, less than a century ago,they were very state of the art.
